Relation Between Heart Beat Fluctuations and Cyclic Alternating Pattern During Sleep in Insomnia Patients. 0
R. de León-Lomelí, J.S. Murguía, I. Chouvarda, IEEE Member, MO Mendez, IEEE Member, E. González-Galván, A. Alba, EMBS Member, G. Milioli, A. Grassi, MG. Terzano, L. Parrino.
Abstract— Insomnia is a condition that affects the nervous and muscular system. Thirty percent of the population between 18 and 60 years suffers from insomnia. The effects of this disorder involve problems such as poor school or job performance and traffic accidents. In addition, patients with insomnia present changes in the cardiac function during sleep. Furthermore, the structure of electroencephalographic Aphases, which builds up the Cyclic Alternating Pattern during sleep, is related to the insomnia events. Therefore, the relationship between these brain activations (A-phases) and the autonomic nervous system would be of interest, revealing the interplay of central and autonomic activity during insomnia. With this goal, a study of the relationship between A-phases and heart rate fluctuations is presented. Polysomnography recording of five healthy subjects, five sleep misperception patients and five patients with psychophysiological insomnia were used in the study. Detrended Fluctuation Analysis (DFA) was used in order to evaluate the heart rate dynamics and this was correlated with the number of A-phases. The results suggest that pathological patients present changes in the dynamics of the heart rate. This is reflected in the modification of A-phases dynamics, which seems to modify of heart rate dynamics.
I. INTRODUCTION Insomnia is a condition where the subject experiences an inability to sleep, which prevent the human body from resting, and therefore affects the nervous and muscular system. The effects of this disorder often involve problems such as poor school or job performance, traffic accidents and also are related with cardiac fails [1]. In this paper we consider two types of primary insomnia: psychophysiological insomnia and paradoxical insomnia which are compared with subjects without sleep disorders [2].
Research supported by SEP-CONACyT grant CB-2010 154623 and CB2012 180604. R. de León- Lomelí, is with the Universidad Autónoma de San Luis Potosí México were she is currently working toward the PhD degree. email: roxy_dl@ hotmail.com. I. Chouvarda is at the Lab of Medical Informatics, Aristotle University of Thessaloniki, Greece. M. Mendez, A, S.Murgía, A. Alba, and E. González are with the Universidad Autónoma de San Luis Potosí, Diagonal Sur S/N, Zona Universitaria, San Luis Potosí, S.L.P., México L. Parrino, A. Grassi, and M. Terzano are at Sleep Disorders Centre, Department of Neurology, University of Parma, Parma, Italy.
978-1-4244-7929-0/14/$26.00 ©2014 IEEE
Previous studies showed that insomnia has a relationship with CAP (Cyclic Alternating Pattern) [3]. CAP is a periodic activity reflected in the EEG during non-REM sleep, and is characterized by sequences of transient electrocortical events, called A-phases, deviating from the baseline of the EEG activity, and occurs repeatedly in 2 to 60 seconds interval. Increased CAP (generally, also the number of Aphases increase) itself may indicate unstable sleep and / or sleep disturbance, and can also be used to identify patterns associated with some sleep disorders. A-phases are identified as A1, A2 and A3, and are closely related to the stages of sleep wakefulness and NREM [2]. A-phases are classified in three groups based on the observed frequency information: i. A1-phase. It is characterized by bursts and k-complexes of Delta waves (0.5 Hz - 4 Hz). ii. A2-phase. It has rapid EEG waves that cover between 20% and 50% of the A-phase duration. iii. A3-phase. It is characterized by Alpha (8 Hz - 12 Hz) and Beta waves (12 Hz - 30 Hz), which cover more than 50% of the A-phase duration [3]. Moreover, the dynamics of the heart rate could be characterized by the Heart Rate Variability (HRV), computed as the time between the detection of consecutive R peaks, obtained from an electrocardiogram signal [4]. Previous analysis focused on the HRV signals in apnea patients and other pathologies, finding a correlation grade between them, using DFA as analysis method. However the correlation between A-phases and heart dynamics in sleep stages has not been analyzed before. And even there is little information about the interaction of the time series in HRV and EEG sleep periods [4] [5] [6]. The main goal of this study is to investigate potential correlation between the A-phases and the HRV signals and to understand the effect of insomnia in them. DFA analysis of HRV is applied and its changes with sleep macro and icro structure are discussed. II. METHODS AND MATERIALS A. Mathematical analysis In 1994 C. K. Peng et al. [7] proposed a method, referred to as Detrended Fluctuation Analysis (DFA) that enables the
2249
detection of long-range correlations in non-stationary time series. This method uses a coefficient (α) to determine the complexity of the signals. A typical example is the R-R series.
y (ti ) be a time series, where ti it and i 1,..., N with a sample rate ∆t. The DFA algorithm is Let
computed as follows [8]: I. Determine the profile
x(ti ) into windows of length n, corresponding to a time scale nt .
II. Divide the integrated time series
III. Calculate the local trend for each window by a leastsquares polynomial fit of the integrated series. The interpolated curve x pol ,m (ti , ) represents the local trend at each window, where m is the degree of the polynomial function. IV. Compute the local fluctuation sequence associated to each window, which is given by
zm (ti , ) x(ti ) x pol , m (ti , )
(2)
For i 1,..., N. V. Calculate the fluctuation function Fm ( ) defined as
1 N Fm ( ) zm (t j , ) 2 N j 1
(3)
which corresponds to the root mean square of the sequence zm (ti , ) .
presents two slopes α1 and α2. Some references indicate that this occurs in 75% of cases and only the first slope helps to determine the level of signal correlation and so the level of patient pathology [7] [8]. This paper focuses on the analysis of the signals of patients with insomnia problems based only on the result of α1. As an example, Fig. 1 illustrates the application of the DFA procedure to a frame of HRV data at an insomnia patient. This Figure only shows the procedure for the case of a window size of 37 points. B. Data analysis The study was carried out on three different groups of patients: (a) normal sleepers (Nor group), (b) psychophysiological insomnia group (PsI group), and (c) sleep misperception group (Mis group). The three groups considered five subjects, two males and three females, where one sleep polysomnographic recording per subject was provided by the Parma University Sleep Disorders Center [2]. In the Nor group the subjects were healthy with a mean age of 36 years and no sleep complaints. The subjects in the PsI group suffered psychophysiological insomnia, with a mean age of 41.5 years. On the other hand, in the Mis group the subjects have paradoxical insomnia, also known as sleep state misperception, with a mean age of 36.2 years.
Repeat the above procedure for a wide range of segments of length n. According to the author recommendations (Penzel [7]), the size of windows should be between nmin 5 and nmax N / 4.
2.5
Original signal Integrated signal Trend Detrended signal
0.12 0.1
2
2.5
2.5
2
2
0.08 1.5
1
0.06
1.5
0.04 0.02 1
0 -0.02
In order to determine if the analyzed signal presents a scaling behavior, the fluctuation function Fm ( ) should reveal a power law scaling
Fm ( ) m ,
(4)
where the scaling exponent α m can be estimated as the slope of the line in a log( Fm ( )) versus log( ) plot, and it quantifies the correlation properties of the time series. Here, we employ a linear polynomial in the detrending procedure, thus m=1 , and for convenience we denote 1 .
2250
0.5
0.5
-0.04
1.5
1
Detrended Signal
(1) of the time series, which is the cumulative sum of the time series from which the series mean value is subtracted.
Integrated signal
i 1,..., N ,
j 1
Original signal
x(ti ) y (t j ) y ,
An interesting phenomenon that some signals may display is the crossover, when the log-log graph of Fm (τ)
Trend
i
The value of α=1/2 corresponds to an uncorrelated signal, and it presents a white noise behavior. On the other hand, if 0
R. de León-Lomelí, J.S. Murguía, I. Chouvarda, IEEE Member, MO Mendez, IEEE Member, E. González-Galván, A. Alba, EMBS Member, G. Milioli, A. Grassi, MG. Terzano, L. Parrino.
Abstract— Insomnia is a condition that affects the nervous and muscular system. Thirty percent of the population between 18 and 60 years suffers from insomnia. The effects of this disorder involve problems such as poor school or job performance and traffic accidents. In addition, patients with insomnia present changes in the cardiac function during sleep. Furthermore, the structure of electroencephalographic Aphases, which builds up the Cyclic Alternating Pattern during sleep, is related to the insomnia events. Therefore, the relationship between these brain activations (A-phases) and the autonomic nervous system would be of interest, revealing the interplay of central and autonomic activity during insomnia. With this goal, a study of the relationship between A-phases and heart rate fluctuations is presented. Polysomnography recording of five healthy subjects, five sleep misperception patients and five patients with psychophysiological insomnia were used in the study. Detrended Fluctuation Analysis (DFA) was used in order to evaluate the heart rate dynamics and this was correlated with the number of A-phases. The results suggest that pathological patients present changes in the dynamics of the heart rate. This is reflected in the modification of A-phases dynamics, which seems to modify of heart rate dynamics.
I. INTRODUCTION Insomnia is a condition where the subject experiences an inability to sleep, which prevent the human body from resting, and therefore affects the nervous and muscular system. The effects of this disorder often involve problems such as poor school or job performance, traffic accidents and also are related with cardiac fails [1]. In this paper we consider two types of primary insomnia: psychophysiological insomnia and paradoxical insomnia which are compared with subjects without sleep disorders [2].
Research supported by SEP-CONACyT grant CB-2010 154623 and CB2012 180604. R. de León- Lomelí, is with the Universidad Autónoma de San Luis Potosí México were she is currently working toward the PhD degree. email: roxy_dl@ hotmail.com. I. Chouvarda is at the Lab of Medical Informatics, Aristotle University of Thessaloniki, Greece. M. Mendez, A, S.Murgía, A. Alba, and E. González are with the Universidad Autónoma de San Luis Potosí, Diagonal Sur S/N, Zona Universitaria, San Luis Potosí, S.L.P., México L. Parrino, A. Grassi, and M. Terzano are at Sleep Disorders Centre, Department of Neurology, University of Parma, Parma, Italy.
978-1-4244-7929-0/14/$26.00 ©2014 IEEE
Previous studies showed that insomnia has a relationship with CAP (Cyclic Alternating Pattern) [3]. CAP is a periodic activity reflected in the EEG during non-REM sleep, and is characterized by sequences of transient electrocortical events, called A-phases, deviating from the baseline of the EEG activity, and occurs repeatedly in 2 to 60 seconds interval. Increased CAP (generally, also the number of Aphases increase) itself may indicate unstable sleep and / or sleep disturbance, and can also be used to identify patterns associated with some sleep disorders. A-phases are identified as A1, A2 and A3, and are closely related to the stages of sleep wakefulness and NREM [2]. A-phases are classified in three groups based on the observed frequency information: i. A1-phase. It is characterized by bursts and k-complexes of Delta waves (0.5 Hz - 4 Hz). ii. A2-phase. It has rapid EEG waves that cover between 20% and 50% of the A-phase duration. iii. A3-phase. It is characterized by Alpha (8 Hz - 12 Hz) and Beta waves (12 Hz - 30 Hz), which cover more than 50% of the A-phase duration [3]. Moreover, the dynamics of the heart rate could be characterized by the Heart Rate Variability (HRV), computed as the time between the detection of consecutive R peaks, obtained from an electrocardiogram signal [4]. Previous analysis focused on the HRV signals in apnea patients and other pathologies, finding a correlation grade between them, using DFA as analysis method. However the correlation between A-phases and heart dynamics in sleep stages has not been analyzed before. And even there is little information about the interaction of the time series in HRV and EEG sleep periods [4] [5] [6]. The main goal of this study is to investigate potential correlation between the A-phases and the HRV signals and to understand the effect of insomnia in them. DFA analysis of HRV is applied and its changes with sleep macro and icro structure are discussed. II. METHODS AND MATERIALS A. Mathematical analysis In 1994 C. K. Peng et al. [7] proposed a method, referred to as Detrended Fluctuation Analysis (DFA) that enables the
2249
detection of long-range correlations in non-stationary time series. This method uses a coefficient (α) to determine the complexity of the signals. A typical example is the R-R series.
y (ti ) be a time series, where ti it and i 1,..., N with a sample rate ∆t. The DFA algorithm is Let
computed as follows [8]: I. Determine the profile
x(ti ) into windows of length n, corresponding to a time scale nt .
II. Divide the integrated time series
III. Calculate the local trend for each window by a leastsquares polynomial fit of the integrated series. The interpolated curve x pol ,m (ti , ) represents the local trend at each window, where m is the degree of the polynomial function. IV. Compute the local fluctuation sequence associated to each window, which is given by
zm (ti , ) x(ti ) x pol , m (ti , )
(2)
For i 1,..., N. V. Calculate the fluctuation function Fm ( ) defined as
1 N Fm ( ) zm (t j , ) 2 N j 1
(3)
which corresponds to the root mean square of the sequence zm (ti , ) .
presents two slopes α1 and α2. Some references indicate that this occurs in 75% of cases and only the first slope helps to determine the level of signal correlation and so the level of patient pathology [7] [8]. This paper focuses on the analysis of the signals of patients with insomnia problems based only on the result of α1. As an example, Fig. 1 illustrates the application of the DFA procedure to a frame of HRV data at an insomnia patient. This Figure only shows the procedure for the case of a window size of 37 points. B. Data analysis The study was carried out on three different groups of patients: (a) normal sleepers (Nor group), (b) psychophysiological insomnia group (PsI group), and (c) sleep misperception group (Mis group). The three groups considered five subjects, two males and three females, where one sleep polysomnographic recording per subject was provided by the Parma University Sleep Disorders Center [2]. In the Nor group the subjects were healthy with a mean age of 36 years and no sleep complaints. The subjects in the PsI group suffered psychophysiological insomnia, with a mean age of 41.5 years. On the other hand, in the Mis group the subjects have paradoxical insomnia, also known as sleep state misperception, with a mean age of 36.2 years.
Repeat the above procedure for a wide range of segments of length n. According to the author recommendations (Penzel [7]), the size of windows should be between nmin 5 and nmax N / 4.
2.5
Original signal Integrated signal Trend Detrended signal
0.12 0.1
2
2.5
2.5
2
2
0.08 1.5
1
0.06
1.5
0.04 0.02 1
0 -0.02
In order to determine if the analyzed signal presents a scaling behavior, the fluctuation function Fm ( ) should reveal a power law scaling
Fm ( ) m ,
(4)
where the scaling exponent α m can be estimated as the slope of the line in a log( Fm ( )) versus log( ) plot, and it quantifies the correlation properties of the time series. Here, we employ a linear polynomial in the detrending procedure, thus m=1 , and for convenience we denote 1 .
2250
0.5
0.5
-0.04
1.5
1
Detrended Signal
(1) of the time series, which is the cumulative sum of the time series from which the series mean value is subtracted.
Integrated signal
i 1,..., N ,
j 1
Original signal
x(ti ) y (t j ) y ,
An interesting phenomenon that some signals may display is the crossover, when the log-log graph of Fm (τ)
Trend
i
The value of α=1/2 corresponds to an uncorrelated signal, and it presents a white noise behavior. On the other hand, if 0
